Jacqueline Ernest 1, Sandra Grañana Castillo 1, Oleg Stepanov 2, Lindsay Clegg 3, Weifeng Tang 3, Diansong Zhou 4, Pradeep Sharma 2
1 Clinical Pharmacology and Quantitative Pharmacology, Biopharmaceuticals R&D, Astrazeneca (Barcelona, Spain), 2 Clinical Pharmacology and Quantitative Pharmacology, Biopharmaceuticals R&D, Astrazeneca (Cambridge, United Kingdom), 3 Clinical Pharmacology and Quantitative Pharmacology, Biopharmaceuticals R&D, Astrazeneca (Gaithersburg, United States), 4 Clinical Pharmacology and Quantitative Pharmacology, Biopharmaceuticals R&D, Astrazeneca (Waltham, United States)
Objectives:
Physiologically based pharmacokinetic (PBPK) reporting often depends on manual collation of PK parameters across numerous simulation outputs and substantial time to write the report. The process is time‑consuming, repetitive, and susceptible to formatting or transcription errors, creating a substantial burden for the modeler and for quality control (QC). The objective of this work was to develop an R Shiny-based application that automates PBPK output generation including 1) automated figure and table generation including simulated and observed data with typical data transformations and standardized formatting, 2) visualization of modelling data across multiple simulations (trial design, drug input parameters, PK parameters, etc), 3) automated text generation for drafting captions and result text based on figure and table results using a large language model.
Methods:
An R Shiny application was developed to import PBPK simulation outputs from Simcyp (version 24, Certara UK Ltd.) [1] model outputs to the R environment. Automated routines were implemented to parse, organize, and standardize key simulation results into report‑ready tables using R packages shiny, tidyverse, flextable, officer and others. The interface was created such that a folder of simulations could be summarized in a single view. By default, Cmax and AUC are summarized as geometric means and confidence intervals with flexible selection of alternative parameters or metrics by the user. Simulation results can be compared against the observed clinical data to calculate ratios and prediction accuracy matrices and to determine bias (geometric mean fold error, GMFE) and precision (absolute average fold error, AAFE and % prediction error, PPE). A separate tab displays the complete summary statistics for each run by file name. Additional tables can be generated to display the study design across simulations. A chatbot function (default, Ollama llama3.2) was implemented to draft report text. Saving features export results to a folder, and a workflow executable within the application inserts formatted tables and figures into Microsoft Word‑based PBPK report templates. Tool performance was evaluated by comparing automated outputs to manually produced tables across representative PBPK projects, assessing accuracy, reproducibility, and time savings. To exemplify the workflow, a drug-drug interaction project for lamotrigine (substrate) and valproic acid (inhibitor) was used to compare time and accuracy of manual assembly compared to the Shiny application.
Results
The Shiny application integrated seamlessly with standard Simcyp outputs, enabling interactive data exploration, cross‑file comparisons, and automated consistency checks. The application successfully automated the generation of standard PBPK reporting tables across multiple simulation files, reducing manual intervention and standardizing formatting. Automated data ingestion improved reproducibility and increased speed. Simulations of lamotrigine with valproic acid could be accurately summarized including predicted AUC and Cmax ratios with and without the inhibitor. The clinical observed data could be accurately integrated from the Shiny application to determine model accuracy. Using the chatbot function, the user can output preliminary captions for tables and figures and suggest text to describe the results which can be further refined by manual typing or improved with additional prompts. The workflow reduced time from model completion to generation of report tables from days to minutes and reduced QC burden. The interactive interface facilitated efficient integration of figures and tables into reporting documents with no coding necessary.
Conclusions
The Shiny-based application provides an efficient and reproducible solution for PBPK reporting workflows. Automation of table generation and embedded QC capabilities streamline reporting, reduce operational timelines, and improve data integrity. Ongoing development will focus on enhancing user experience and expanding the tool to support broader PBPK modelling needs. The Shiny-based application provides an efficient and reproducible solution for PBPK reporting workflows. Automation of table generation and embedded QC capabilities streamline reporting, reduce operational timelines, and improve data integrity.
References:
[1] Jamei M, Marciniak S, Edwards D, Wragg K, Feng K, Barnett A, Rostami-Hodjegan A. The simcyp population based simulator: architecture, implementation, and quality assurance. In Silico Pharmacol. 2013 Jun 3;1:9. doi: 10.1186/2193-9616-1-9. PMID: 25505654; PMCID: PMC4230310.
Reference: PAGE 34 (2026) Abstr 12006 [www.page-meeting.org/?abstract=12006]
Poster: Drug/Disease Modelling - Absorption & PBPK