Chapter 3 92 ◀ Figure 1. Immunohistochemical analysis of all biomarkers and digital mark-up. (A) The left image shows an example of a CD8 immunohistochemical staining (IHC). The black square indicates the area that is shown in the central image. The right image shows the digital markup with CD8 TILs (in brown) and all other cells (in grey). (B) The left image shows the same example as shown in A. The black square indicates the area that is shown in the central image. The right image shows regional analysis of only intratumoral (IT) CD8 TILs. Stromal CD8 TILs are not quantified. Red lines indicate the tumor region. Red arrows indicate ITCD8 TILs. White arrow indicates the area with stromal CD8 TILs. (C) The left image shows an example of a consecutive slide stained for PD-1 IHC. The black square indicates the area that is shown in the central image. The right image shows the digital markup with PD1T TILs (in brown) and all other cells (in grey). (D) The left image shows an example of a consecutive slide double stained with CD20 and CD3 IHC. The black square indicates the area that is shown in the central image with CD20+ B cells (in yellow) and CD3+ T cells (in purple) localizing in a TLS. The right image shows the digital markup with CD20-positive areas highlighted by the intensity of the yellow staining (depicted as spectrum from yellow to red color). (E) Example of a consecutive slide stained for PD-L1 IHC. The black square indicates the area that is shown in the right image. PD-L1 IHC slides were scored manually. Accuracy of individual and composite biomarkers to predict DC at 6 months Next, optimal cut-off values for individual and composite biomarkers in the training cohort were determined through ROC analysis. We aimed for a sensitivity and NPV of ≥90% to minimize the risk of undertreatment, while maintaining a specificity of at least 50% to identify patients unlikely to respond to PD-1 blockade therapy. The latter group can potentially benefit from alternative treatments. Since not all tumor samples were evaluable for all nine biomarkers, the number of samples in the training cohort ranged from 28 to 55 (Table 1, Fig. S1). A total of 16 composite biomarkers, including PD-1T and TIS as individual biomarkers, met the prespecified sensitivity and specificity criteria on the ROC curve (Table S2). Interestingly, among these, cut-off values of 7/8 (88%) possible combinations with PD-1T TILs and 5/8 (63%) with TIS reached these criteria (Table S2). However, none significantly improved predictive accuracy compared to the standalone use of PD-1T TILs and TIS (Fig. S2A,B), leading to their exclusion from further analysis. Subsequently, we selected the four remaining biomarkers with the highest predictive performance for validation, including the combinations of CD8+ITCD8 and CD3+IT-CD8, as well as the individual biomarkers PD-1T TILs and TIS (Table S2). In the training cohort, both CD8+IT-CD8 and CD3+IT-CD8 demonstrated significantly higher probability scores in the DC 6m group (reflecting the probability of patients reaching DC 6m) compared to the progressive disease (PD) group (CD8+IT-CD8, P<0.0001 and CD3+IT-CD8, P<0.001) (Fig. 2A,B). The area under the ROC curve (AUC) was 0.83 (95% CI: 0.73-0.94) for CD8+IT-CD8, and 0.78
RkJQdWJsaXNoZXIy MTk4NDMw