Programming Director, CPMS Programming and Business Excellence GSK, United States
Disclosure(s):
Xiaobin Li, M.S: No financial relationships to disclose
Objectives: Tasks to explore clinical data, validate pharmacometrics (PMX) datasets and related input data, and conduct exploratory data analysis (EDA) are interconnected and share commonalities. Although many tasks are common and iterative in principle, specific requirements may vary between studies. To meet these requirements, the xplorer shiny app[1] was developed offering a customizable automated approach to explore data effectively. Furthermore, it improves quality assurance while reducing dataset delivery time.
Methods: An R package[2] was developed supporting EDA and shiny app development, enabling usage of well-structured functions. The package covers data viewing, summarizing, graphical analyses, and automated reporting. It incorporates generic and specialized data checks essential for PMX dataset creation. This suite of checks will reveal 1) abnormal data within the source data, 2) potential coding or derived issues in PMX datasets, 3) the relationships between source and analysis data.
The shiny app within the R package creates a user friendly and efficient workflow. Emphasizing on flexibility, the app imposes no restrictions on data standard. All provided functionality can be used for any applicable data type, e.g., ADaM, SDTM or PMX data. Efficiency is enabled by the ability to implement both user specific and global settings. Global settings provide meaningful starting points for tasks, while user setting allows customizing the tasks. More customizations can be achieved by enabling R coding and data manipulation within the app, while also logging scripts for further customization, traceability and reviewing.
For reusability, settings in the app can be adapted, saved, and shared to quickly (re)create results. Settings can also be combined into profiles, enabling batch actions and report generation in various formats. To ensure traceability and reproducibility, critical information including source data file paths, task parameters, associated R scripts, and Rmarkdown reports can be saved.
Results: The R package and accompanying modular shiny app are being released within GSK. Initial usage yielded positive user feedback. The ability to rerun global and user settings as well as profiles reduces time to understand and explore data. The common and domain specific data checks improve data quality and traceability. Effective visualization facilities communications between programmers and modelers.
Conclusions: Implementation of this shiny app designed for data visualization, quality checking, and EDA enables users to make better informed decisions. With a focus on flexibility, reusability, reproducibility, and audit trail functionality, this tool not only accelerates the process of obtaining high-quality PMX datasets but also gains insights on data to initiate model development. By streamlining tedious tasks, the app enhances productivity, reliability, and ultimately, supports the acceleration of overall submission timelines.
