(T-025) Overcoming Temporal and Quality Challenges in Dataset Creation for Pharmacometrics Analysis

Shinri Ikari – Kyowa Kirin Co., Ltd.; Dimple Patel – Kyowa Kirin, Inc.; Yoshinori Nagata – Kyowa Kirin Co., Ltd.; Douglas Marsteller – Kyowa Kirin, Inc.

Objectives: Preparing datasets for PMx analysis can be complex and time-consuming, requiring highly trained data expertise. This is especially true in PPK studies, which integrate data from multiple clinical trials. Each trial may have slight differences in SDTM structure, requiring deep understanding and manual effort for consistency. Additionally, changes in model structure during analysis often require dataset modifications. This initiative aims to enable timely analyses during clinical trials and provide efficient feedback. To achieve this, it is essential to reduce the time required to create analysis datasets and to ensure consistent quality.

Methods: We addressed temporal and quality issues by assigning SDTM programmers to create the integrated dataset in parallel with the SDTM domains. This dataset, which combines relevant SDTM domains, follows a simple vertical structure for the LB domain. The specifications for this integrated dataset were written in a format similar to SDTM specifications. We did not ask SDTM programmers to implement any PMx-specific dataset structures. Instead, we optimized the variable definitions of the integrated dataset so that pharmacometricians could semi-automatically convert the integrated dataset into various model-specific formats using the R package IQRtools to create the analysis dataset..

Results: Based on the devised specifications, SDTM programmers were able to create the integrated dataset with limited time lag from the completion of the SDTM. Pharmacometricians were able to conduct various model analyses using the created integrated dataset. The basic indicators to be included in the integrated dataset were defined, resulting in a structure that did not require significant changes and encompassed trials of different phases and indications. Variable indicators dependent on specific diseases or drugs profiles were made optional. Most of the programs for the dataset, which is based on SDTM and resembles the LB domain, can be templated.

Conclusions: With the development of this specification and improvements in procedure, it is now possible for SDTM programmers to create integrated datasets concurrently with SDTM. The structure of the integrated dataset created through this process is simple, vertical, and independent of trial design. Thus, integrated analysis can be conducted by simply stacking datasets from multiple trials vertically with this efficient and effective programming approach.

Citations: none

Keywords: PMx,
Statistics,
Dataset