As shown in the table below a number of product development items were proactively identified and addressed during the development phase of the assay.
| Table 1: DecisionDx-GBM Development Considerations | |
|---|---|
| Development Considerations | Solution |
| Generation of false positives due multiple comparisons of a single dataset | Use multi-institutional sample sets in discovery effort to identify robust gene set. |
| Specimen platform that is widely available | While there are benefits to using fresh frozen tumor samples for RNA analysis, the individual patient use of this test would be limited if it were dependent upon fresh frozen tissue. Therefore, an early aim of the inventors was to transfer the platform to using formalin-fixed paraffin embedded (FFPE) tissue. This transfer was successful. |
| Improve efficiency | While the discovery efforts identified a robust set of 38 genes that were common among all four of the multi-institutional sample set, efforts were planned to reduce the gene set to the most efficient set possible. We were able to reduce the set to 9 genes of interest (plus 3 controls). |
One of the issues in identifying biomarkers is the potential for generating false positive biomarkers as a result of running multiple comparisons from a single dataset. One of the methods to control for this issue is to use independent sample sets during the discovery process. Sample sets from Massachusetts General Hospital (MGH), ULCA, UCSF and M. D. Anderson Cancer Center were used in the development of this test.
The discovery effort focused on identifying the most robust, top 200, survival genes in each of these sample sets. As shown in the Venn diagram below, 38 genes were robustly associated with all four sample sets. Additional analysis using scrambled survival data showed a false discovery rate estimated at 0.3%. While SAM, Rank Product, an Cox model were all conducted, the largest overlap of these consensus genes was identified through the fold-change method. The fold-change method was used for all subsequent analysis and in the clinical test.
