Serum test identifies patients deriving benefit from CPIs 177 5 made for training classes, with patients with the lowest time-to-event outcome times assigned to the ‘negative’ class and those with the highest time-to-event outcome times assigned to the ‘positive’ class. A classifier is constructed using these training classes and used to generate classifications for all samples in the development set using out-of-bag evaluations. These resulting classifications are then used as improved training class labels for a second iteration of classifier construction. This simultaneous iterative refinement of the classifier and the classes used in training generally converges quickly and reveals the underlying structure of the MS data and its association with clinical outcomes19. Full details of the application of the method in this setting are provided in the supplementary materials. One classifier previously developed with the Diagnostic Cortex platform was used as part of the developed test. BDX008 was created to stratify patients with advanced melanoma into groups with better and worse outcomes when treated with nivolumab15. It has been validated in multiple independent cohorts of melanoma patients treated with CPIs15,20. Also, it has demonstrated some ability to stratify overall survival of patients with advanced NSCLC treated with nivolumab21. A version of BDX008, adapted for the spectral preprocessing parameters and feature definitions in this project, was created (See Supplemental Data: Methods for details). Preliminary statistical considerations showed a binary split of the development set into equal-sized groups would allow detection of a hazard ratio between the groups of 0.55 with 90% power, assuming fully mature clinical data and a significance level of 95%. Similar considerations for a ternary split into equal size subgroups would allow detection of a hazard ratio of 0.48 under the same specifications. All reference data and test parameters were generated solely using the development set. Validation sets and the chemotherapy cohort were never used in the creation of any components of the test. All elements of the classification algorithms were locked prior to running the test on the validation sets and chemotherapy cohort. Protein set enrichment analysis (PSEA) This analysis applies the gene set enrichment analysis (GSEA) method22 to protein expression data. The method identifies expression differences that are consistent across prespecified groups or sets of attributes, in this case, sets of proteins that are associated with particular biological processes. Two additional independent reference sets of serum samples with matched MS data and protein expression data were used for this set enrichment analysis. One sample set was composed of 49 samples with protein expression data from a panel of 1,129 proteins; the second
RkJQdWJsaXNoZXIy MTk4NDMw