Software and Data Standardization and Integrity
Chair: Fang Chen (SAS)
Vice Chair: Shibing Deng (Pfizer)
Speaker: Kaifeng Lu (BeiGene)
Title: Introducing the lrstat package
Abstract: Delayed treatment effect is a common phenomenon observed in immuno oncology clinical trials. Although there is a rich literature on the power and sample size calculations for log rank test under non proportional hazards, most statistical software either deals with non-proportional hazards for the fixed design only or resorts to simulations for non-proportional hazards in general. The lrstat package intends to fill the gap for analytical power and sample size calculations for log-rank test under non proportional hazards. It also provides simulation tools to verify the analytical calculations. In addition, the package contains useful functions for designing a generic group sequential trial and functions for performing multiplicity adjustments in the group sequential setting using the graphical approach. I will demonstrate the flexibility of the package through an example for sample size calculations under delayed treatment effect and an example of power calculation with multiplicity adjustment for a three arm group sequential trial with two survival endpoints.
Speaker: Douglas M. Robinson (Pfizer, Inc.)
Title: Building tools to drive automation and decision making with clinical biomarkers
Abstract: Technological advances are allowing researchers to collect massive amounts of biomarker data from patients in our clinical trials with minimal sample requirements. This includes Whole Exome Sequencing, Whole Transcriptome Sequencing, Circulating Tumor DNA, among many others. These data are typically much larger than your typical clinical trial data which negatively impacts data ingestion, data migration, data pre-processing and implementing analytical methods. With numerous clinical trials across different therapeutic areas, generating results quickly and efficiently becomes critical to impact clinical trial team decisions in a timely manner. To do so, analysts must be able to present these complex biomarker data in an easily digestible format. In this presentation, we will explain how Pfizer provides robust statistical analyses of its clinical biomarker data and provide insight into a novel initiative to enhance turnaround times on a first pass analysis of biomarker data.
Speaker: Huiping Miao (SAS)
Title: Software Validation at SAS
Abstract: This presentation provides an overview of software testing at SAS, with a focus on validation of statistical software products such as SAS/STAT. Productization of commercially built software from SAS involves heavy testing before delivery to customers. Ensuring that the software works as documented, that it is backward compatible, that it works properly in various operating systems, configurations, releases, distributed environments, and security environments are of essential importance in this process. An additive and substantial component of the testing specifically for analytics software is to ensure numerical accuracy.
Test engineers build myriad suites of test programs and benchmark the results, including numerical outputs, messages, return codes, and graphical displays. A thorough test suite detects unexpected side effects, often on different platforms, due to changes or enhancements made to the software. Validation is performed in a variety of ways, which can include independent numerical implementation of statistical models using identical solution algorithms and convergence criteria, simulations, verification against results in publications, stress testing, and consistency testing. Examples of the testing process and tools used at SAS will be shown.
