CTS - Clinical Trials Simulation System
Peter H. Westfall1, Kuenhi Tsai2, Miles Dunn2, Stephan Ogenstad3,
Alin Tomoiaga1, Yongang Lu1, Keyi Wang1
1 – Texas Tech University
2 – Vertex Pharmaceuticals
3 – Statogen Consulting LLC / iClinTec
I. Overview
This is a general trial generation/analysis system. It allows the user to input parameters of a clinical trial, and it generates many hypothetical trials resulting from such input specifications.
The trials have a “typical” form, with multiple records per patient data from repeat visits to the treatment center, with multiple treatment groups and, as well as patient dropouts. These simulated trials can then be used to design trials optimally for sample size allocation, patient length on trial, choice of statistical test, etc., potentially saving costs associated with inefficient trial and protocol design.
The data sets are huge and the analyses are complex, so supercomputing is recommended. This system can be run using a grid or locally, producing “canned” and customizable output.
Parameters are entered using a SAS/AF graphical user interface and can be saved for ease of future use. There are ~3,100 lines SCL code; ~2,300 lines of SAS/BASE/STAT/Macro code,
and hundreds of SAS/AF instructions, all embedded in the system.
The main approach is statistical; Pharmocokinetic/Pharmacodynamic (PK/PD) models are used only indirectly. The goal is to simulate realistic data sets with appropriate mean, covariance, and distributional structures that are needed for answering statistical questions in protocol and trial design, with emphasis on Phase II/III trials, but which can be customized for certain Phase I designs (eg crossover studies) as well. The system generates multiple timepoint/endpoint data with flexible covariance structures, flexible mean structures, including natural history and placebo effects, flexible distributions including survival, compliance effects, and informative dropout mechanisms.