Background: Anti-PD1 and PD-L1 antibodies (mAb) are immune checkpoint inhibitors (ICIs) to treat patients with metastatic non–small cell lung cancer (NSCLC). Unfortunately, only a handful of patients respond to ICIs. Methods: A cohort of patients with metastatic NSCLC (n=133) treated with anti-PD1 or anti-PD-L1 mAb in two independent care centers was evaluated. An independent cohort of 132 patients from another hospital was used as a validation. Immunoscore IC, an in vitro diagnostic test (CE-IVD), was used on a routine single FFPE slide, and duplex immunohistochemistry CD8 and PD-L1 staining was quantified using digital pathology. Quantitative and spatial parameters related to cell location, number, proximity, and clustering were analyzed. An Immunoscore IC–based model discriminated patients into 2 categories or 3 categories. Results: Anti–PD-L1 clone (HDX3) had similar characteristics as other anti–PD-L1 clones (22C3, SP263) with a mean overall agreement above 95%. Intra- and inter-laboratory concordances for classifying patients at 1% cut-off according to digital anti–PD-L1 (HDX3) were 100% and 94%, respectively. Routine laboratory evaluation of PD-L1 expression showed an agreement with digital anti–PD-L1 quantification of 92% and 97% at 1% and 50% cut-offs, respectively. Using univariate Cox model after FDR correction, 5 pathological dichotomized variables were significantly associated with PFS (all p < 0.0001). These variables included: CD8 free of PD-L1, CD8 clusters, CD8 cells in proximity of PD-L1 cells, CD8 number, PD-L1 cells in proximity of CD8 cells. Similar results were found using univariate Cox analysis on continuous variables (all p < 0.003) in two independent cohorts of patients. Using multivariate Cox model Immunoscore IC classification improved the discriminating power of prognostic model, which included clinical variables and pathologist PD-L1 assessment. In two categories, the Immunoscore IC risk score was significantly associated with both patients’ PFS (p < 0.0001) and OS (p < 0.0001) in the training cohort and in the validation cohort (PFS: p = 0.0047, OS: p < 0.0001). Further increased hazard ratios were found when stratifying patients into 3 categories of Immunoscore IC. At 6 months, PFS rates were 10% versus 60% in the training cohort and 20% versus 62% in the validation cohort for high-risk and low-risk Immunoscore IC score, respectively. All patients (100%) with high-risk Immunoscore IC score relapsed in less than 18 months, in contrast to 34% and 33% of low-risk Immunoscore IC patients who did not relapse for more than 36 months in the training and validation cohorts, respectively. Conclusions: These data underline that Immunoscore IC is a potent tool to predict the efficacy of ICIs in patients with NSCLC. Immunoscore IC characterized patients who are resistant to ICIs.