Transversal Learning in Regulatory application of Complex Mechanistic models in Risk Assessment. Thyroid Disruptors and what we learn from Drug Development

Stephan Schaller¹, Alexander Kulesza¹, Pavel Balazki¹

¹ESQlabs GmbH, Saterland, Germany

Background: Next-Generation Risk Assessment (NGRA) integrates mechanistic models with in vitro and in vivo data to reduce animal testing and improve human relevance. Transparent uncertainty characterization across evidence streams is a critical regulatory requirement, formalized in the OECD Guidance on Physiologically Based Kinetic (PBK) models. Bayesian hierarchical parameter estimation has proven effective in Model-Informed Drug Development (MIDD) for propagating uncertainty from in vitro assays through preclinical and clinical data, but remains underutilized in chemical risk assessment.

Objective: We explore how hierarchical Bayesian methods from pharmaceuticals could be translated to chemicals, using thyroid hormone disruptors as a conceptual case study. Thyroid disruption involves complex species-specific differences in feedback regulation (TSH), clearance (DIO, UGT), and protein binding (TBG, TTR, albumin), making it an ideal test case for thinking through mechanistic uncertainty propagation in NGRA.

Conceptual Framework: We have developed a modular PBK-QST platform (OSP Suite v12) integrating compound PBK models, thyroid regulation modules, and quantitative adverse outcome pathways (qAOPs). Drawing on learnings from pharmaceutical applications, we propose how a Bayesian multi-level modeling (BMLM) framework could be applied to this platform. The approach would separate global parameters (compound properties, IVIVE hyperparameters for UGT induction, baseline physiology) from individual level variability, propagating uncertainty from in vitro induction assays and rat studies to human T4/T3/TSH predictions. A unified data-alignment layer would support both parameter estimation and OECD-compliant reporting.

Transversal Insights & Requirements: We discuss key methodological considerations when adapting hierarchical Bayes from pharma to chemicals: (i) managing mechanistic complexity and feedback loops in endocrine systems; (ii) addressing data sparsity and identifiability challenges in toxicology; (iii) aligning the proposed workflow to OECD PBK Guidance (characterization, validation, uncertainty analysis); and (iv) integrating posterior distributions into Weight-of-Evidence frameworks for regulatory decision-making.

Conclusions: Adapting proven MIDD hierarchical methods to NGRA could accelerate regulatory confidence in mechanistic chemical risk assessment. This transversal learning would foster a shared, auditable, and extensible modeling ecosystem bridging pharmaceutical and toxicological domains