Composite versus individual biomarkers for predicting clinical benefit to PD-1 blockade in NSCLC 89 3 were made using the McNemar test. A point on the ROC curve corresponding to 90% sensitivity for DC 6m or DC 12m was selected to determine specificity, NPV and PPV. The aim was to achieve an NPV of ≥90% and a specificity of ≥50%. Two (closely related) non-parametric approaches were considered to obtain 90% sensitivity for predicting DC 6m and DC 12m from two biomarkers. In both methods, a cut-point was chosen for each of the two biomarkers and a patient was predicted positive (i.e. likely to respond to PD-1 blockade) if at least one (first method) or both (second method) biomarker values were above their respective cut-point values. The specificities obtained with these non-parametric methods were either equal to or worse than those obtained by the parametric method described above (i.e. via logistic regression). Therefore, these non-parametric methods were not used in this study. Four training models were selected based on a cut-off that demonstrated the highest specificity and NPV at the predefined sensitivities for predicting DC 6m and DC 12m. This cut-off was then used to determine sensitivity, specificity, NPV and PPV in the validation cohort.
RkJQdWJsaXNoZXIy MTk4NDMw