Serum test identifies patients deriving benefit from CPIs 179 5 Results Patient characteristics and overall outcomes for all four cohorts are summarized in Table 1 and were typical of patients with advanced NSCLC treated predominantly in the second-line setting. Clinicopathological characteristics were generally similar between the four cohorts, although the proportion of patients with performance status 2 or higher was larger in the development cohort and validation set 1, and the proportion of patients with performance status 0 was higher in the chemotherapy cohort. PD-L1 status was not available for the chemotherapy cohort and was missing for at least one-third of patients in the other three cohorts. Development of the test A ternary test was developed that was able to stratify the development set of 116 samples into three groups with different outcomes after anti-PD-1 treatment, i.e., the resistant group (with poor outcomes), the intermediate group (with intermediate outcomes) and the sensitive group (with good outcomes). The ternary test result was generated by combining the results of three binary classification algorithms (classifiers). Each of the three classifiers stratified patients into two groups: “positive”, with better outcomes and “negative” with worse outcomes. The binary results were integrated, as shown in figure 1, to yield the final test result. First, classifications were generated for all samples by Classifier A, the version of the pre-existing BDX008 test adapted to the spectral processing used in this project. To identify a group of patients least likely to have good outcomes, the patients classified as negative by Classifier A were subsequently classified by Classifier C. This classifier was developed using the subset of MS features found to be associated with Immune Response type 2 by PSEA and a subset of the development cohort enriched for inferior outcomes, by excluding patients designated as BDX008+ and having performance status 0. (The MS features in this subset are listed in the supplementary materials.) Samples designated as negative by both Classifier A and Classifier C were classified as “resistant”. To identify a group of patients likely to have the best outcomes, the patients classified as positive by Classifier A were further classified by Classifier B. This classifier was developed using all 274 mass spectral features on a subset of the development set enriched for better outcomes, by excluding patients who were classified both as BDX008- and negative by Classifier C. Samples designated positive by both Classifier A and Classifier B were classified as “sensitive”. All samples not classified as “sensitive” or “resistant” were classified as “intermediate”. More details of the test development process and parameters are provided in the supplement. Reproducibility was assessed by running the test on the 98 serum samples of validation set 1 twice, 13 months apart. Concordance between classifications was 85%.
RkJQdWJsaXNoZXIy MTk4NDMw