Background: The optimal duration of immune checkpoint inhibitor (ICI) treatment for patients with advanced non-small cell lung cancer (NSCLC) remains to be determined. Treatment durations in cornerstone phase 3 clinical trials vary between a fixed 2-year duration and pursuit until disease progression. Clinical practices may thus differ according to the attending physician. Objectives: Here we provide real-world data about treatment decisions at 2 years, with subsequent clinical outcomes. Design and Methods: This multicentric observational study included patients with advanced NSCLC whose disease was controlled after 2 years of pembrolizumab or nivolumab. The primary outcome was the decision to discontinue ICI treatment or not, along with factors motivating this decision. Secondary outcomes included progression-free survival (PFS) (according to treatment continuation or not) and adverse events. Results: A total of 91 patients were included, of which 60 (66%) had been pre-treated. The programmed death-ligand 1 expression level was ⩾50% in 43 patients (47%). In 61 patients (67%), ICI was continued after 2 years of treatment. This decision was significantly associated with the care center (p < 0.001) but neither with the tumor response at 2 years, as evaluated by CT scan or PET scan, nor with clinical status, immune-related adverse events, or previous locally treated oligo-progressive disease under ICI. Two years after the 2-year decision, PFS was 68.5%, [95% confidence interval (CI) (53.3–88.0)] in the ‘ICI discontinuation’ group and 64.1% [95% CI (51.9–79.2)] in the ‘ICI pursuit’ group; hazard ratio for relapse was 1.14 [95% CI (0.54–2.30), p = 0.77]. The overall survival rate at 24 months after discontinuation was 89.2% [95% CI (78.4–100)] for the ‘discontinuation’ group and 93.1% [95% CI (85.8–100)] for the ‘pursuit’ group. Given insufficient power, overall survival could not be compared. Conclusion: The decision to continue ICI or not after 2 years of treatment depends mainly on the care center and does not seem to impact survival. Larger, randomized data sets are required to confirm this result.