Chapter 3 88 were not quantified as no reliable algorithm could be established due to dense clustering of CD20+ or CD3+ cells within and at the border of TLS. Tumor areas were digitally annotated as described previously25. CD20 data from 91 samples were used from previous work25 (Table S1). PD-L1 scoring PD-L1 TPS was determined using the qualitative, clinical grade, laboratory developed, IHC assay (22C3 Agilent/DAKO) as described previously25. PD-L1 TPS data from 94 samples were used from previous work25 (Table S1). The CD68 staining was compared to the PD-L1 staining to exclude macrophages that express both CD68 and PD-L1, as their presence could potentially introduce false‑positive results. RNA extraction and hybridization to nCounter tagset The extraction of RNA from pretreatment FFPE samples and subsequent Nanostring analysis were performed as described previously35. Statistical analysis Patient characteristics were descriptively reported using mean ± SD, interquartile range (IQR) or frequencies (percentages). The Mann-Whitney test for continuous data, Fisher’s exact test for categorical data and linear-by-linear association test for ordinal data were used to assess differences in patient characteristics between cohorts (training and validation) and between outcome groups (DC vs PD). Statistical significance was considered at *P<0.05, **P<0.01, ***P<0.001 or ****P<0.0001. Genes in the Tumor Inflammation Signature (TIS) are normalized using a ratio of the expression value to the geometric mean of the housekeeper genes specific to the TIS signature and then followed by log2 transformation. The TIS score, a weighted linear combination of the 18 gene expression values, was calculated as part of Nanostring’s intellectual property29,36 (Table S1). In the training cohort, univariate and bivariate logistic models were constructed for DC 6m and DC 12m using CD8 TILs, IT-CD8 T cells, PD-1T TILs, CD3 TILs, TLS, TLS+LA, CD20+ B cells, PD-L1 and TIS. The bivariate models included an interaction term. The bivariate logistic model produces for each patient a number between 0 and 1, reflecting the probability (according to the model) of patients reaching DC 6m or DC 12m. Discriminatory ability was evaluated using the area under the receiver operating characteristic (ROC) curve. Predictive performance metrics (sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV)) for different individual and composite biomarkers were calculated and comparisons
RkJQdWJsaXNoZXIy MTk4NDMw