Thaddeus H Grasela (1), Sébastien Bihorel (1), Cynthia A Walawander (1), Jill Fiedler-Kelly (1), David Fox (1), and Andrew Rokitka (1)
Cognigen Corporation, a subsidiary of Simulations Plus, Inc, Buffalo, New York
Background: Technical reports for pharmacometric modeling provide comprehensive documentation, typically including data assembly methods and disposition, modeling strategy, and analysis results, but are costly and time consuming and do not necessarily serve the dynamic R&D lifecycle which requires an ongoing accretion of data from Phase 1, 2, and 3 trials. Cognigen implemented KIWI™ in 2011, a secure internet-based service providing high throughput NONMEM® processing. KIWI is the basis for a progressive reporting process that facilitates the capture of critical information during model development, enables team access to evolving, interim results, and facilitates rapid assembly of technical reports.
Objective: Perform an analysis of KIWI metadata to assess system performance and evaluate its use in facilitating progressive reporting.
Methods: NONMEM runs are submitted via KIWI and maintained, along with the analysis datasets, on a secure server. Processing is done by a dedicated cluster of Linux servers. KIWI metadata is stored using a relational database management system, broadly grouped into categories of project management and NONMEM run-specific information. KIWI connection flags denote the use of a specific run for other activities, such as performing a visual predictive check, and can be used for critical path identification.
Results: Modeling efforts for 32 drugs were performed using KIWI over 3.5 years. The mean (SD) duration of a modeling effort was 367 (313) days with 1.75 (1.16) modeling projects per effort and a duration of 214 (227) days per project. About 30 (16) runs per project were labeled as critical path, representing ~2% of the runs per project. Generally, ≥ 10 plots of various types were produced per critical path run and sometimes considerably more. Approximately 80% of covariate analyses required ≤ 3 forward selection steps and 77% required ≤ 3 backward elimination steps. Final models included 15 (16) thetas, 5 (6) etas, 1 (1) off-diagonal eta, and 2 (2) epsilons. Graphs and pre-formatted, report-quality tables of results were easily exported, saving upwards of 2 hours per run.
Conclusions: KIWI provides ready access to analysis metadata that can be used to monitor system requirements and analysis status, as well as forecast resource needs for subsequent modeling efforts. Ongoing efforts are directed at leveraging metadata and run connections to automate progressive reporting and further facilitate the preparation of technical reports.
Reference: PAGE 24 () Abstr 3615 [www.page-meeting.org/?abstract=3615]
Poster: Methodology - Other topics