Composite versus individual biomarkers for predicting clinical benefit to PD-1 blockade in NSCLC 99 3 Training (Continued) Validation Sensitivity Specificity NPV PPV Samples (n) AUC (95%-CI) Sensitivity Specificity NPV PPV 92% 67% 95% 52% 61 0.72 (0.57-0.87) 72% 74% 86% 54% 100% 55% 100% 47% 40 0.57 (0.36-0.77) 83% 39% 84% 37% 94% 62% 96% 50% 77 0.62 (0.50-0.75) 64% 56% 76% 41% 94% 54% 95% 47% 74 0.68 (0.55-0.80) 83% 46% 85% 43% 100% 64% 100% 43% 61 0.80 (0.65-0.94) 86% 74% 95% 50% 100% 50% 100% 35% 40 0.63 (0.43-0.82) 100% 39% 100% 26% 92% 63% 96% 41% 77 0.67 (0.53-0.81) 68% 57% 88% 30% 100% 68% 100% 46% 38 0.59 (0.36-0.82) 29% 68% 81% 17% alone. Despite not reaching statistical significance, possibly due to the low sample size (P=0.34), this combination was selected for further analysis (Fig. S3B, Table S3). Other selected biomarkers for validation with the highest predictive performance included CD8+IT-CD8, PD-1T TILs and TIS (Table S3). The probability scores for DC 12m and PD are depicted per sample in Figure S3C (CD8+IT-CD8, P<0.001) and Figure S3D (CD8+TIS, P<0.01). The two composite biomarkers showed a high AUC of 0.85 (95% CI: 0.73-0.96) (CD8+IT-CD8) and 0.91 (95% CI: 0.79-1.00) (CD8+TIS) in the training cohort, and optimal cut-off values of 0.122 and 0.124, respectively, were chosen (Fig. 3A,B and Table 3). In the validation cohort, only CD8+IT-CD8 showed borderline significantly higher probability scores in the DC 12m group versus the PD group (P=0.03) (Fig. S3E,F). The ROCs resulted in low AUCs (CD8+IT-CD8: 0.67; 95% CI: 0.53-0.81), CD8+TIS: 0.59; 95% CI 0.36-0.81) (Fig. 3A,B). Furthermore, the sensitivity (68% and 29%), specificity (57% and 68%), NPV (88% and 81%) and PPV (30% and 17%) did not meet the prespecified performance criteria (Table 3).
RkJQdWJsaXNoZXIy MTk4NDMw