# From Population-Based PBPK to Individualized Virtual Twins: Clinical Validation and Applications in Medicine

## TL;DR

This paper reviews how PBPK models are shifting from population-based to individualized virtual twins to improve personalized medicine and drug dosing.

## Contribution

The paper highlights the clinical validation and application of individualized PBPK models as virtual twins in medicine.

## Key findings

- Incorporating individual patient data improves PBPK model performance for personalized dosing.
- Renal function and demographic variables enhance predictive accuracy in PBPK models.
- CYP phenotypic and genotypic data show inconsistent benefits in model personalization.

## Abstract

Physiologically based pharmacokinetic (PBPK) models are widely used in the context of personalized medicine, as they allow for the evaluation of dosing schedules and routes of administration by predicting absorption, distribution, metabolism and excretion (ADME) of drugs in biological systems. Traditionally, PBPK models have been developed and applied at the population level, enabling the characterization of predefined cohorts, which remains limited in supporting true precision dosing. In this review, we explored the increasingly common shift from population-based to individual PBPK modelling, where individuals are modelled as virtual twins (VTs). Through the inclusion of additional patient-specific data, such as demographic, physiological, phenotypic and genotypic information, models can be personalized, moving beyond traditional one-size-fits-all strategies. Overall, incorporating individual patient data (e.g., septic, psychiatric, cardiac, or neonatal populations) improves model performance. Physiological parameters, particularly renal function, show strong potential given their role in drug elimination, while demographic variables enhance predictive accuracy in certain studies. In contrast, the benefits of including cytochrome P450 (CYP) phenotypic and genotypic data remain inconsistent. We further emphasize methodologies used to evaluate model performance, with a focus on clinical validation through comparisons between predicted and observed concentration-time profiles. Key challenges, including limited sample sizes and data availability, that may compromise predictive precision, are also discussed. Finally, we highlight the potential integration of PBPK-based VTs into broader digital twin frameworks as a promising path toward clinical translation, while acknowledging the critical barriers that must be addressed to enable routine clinical implementation.