Background: Tumor mutational burden is an emerging biomarker of response to immune checkpoint inhibitors (ICI), whose clinical adoption is challenging, especially in liquid biopsies. We hypothesized that targeting limited but relevant genetic alterations in plasma cell-free DNA with next generation sequencing (NGS) along with early monitoring may represent a non-invasive approach to predict response to ICI. Methods: Plasma samples from responders (PFS > 6 months) and non-responders (progressive disease at first evaluation) patients collected before nivolumab (second line) initiation and at 1 month were tested using tagged amplicon sequencing of hotspots and coding regions from 36 genes, blinded to clinical outcomes and tumor genotype. Molecular profile of ctDNA, and its early kinetics (1 month) were analyzed as potential early indicators of response. Results: 98 patients were analyzed, of which 86 (39 responders, 47 non-responders) were evaluable for response. Alterations in ctDNA were detectable in 67/86 baseline samples (78%). The detection of a targetable oncogenic driver (5 EGFR, 1 ALK) was associated with progressive disease on the first CT-scan The presence of a PTEN and/or STK11 mutations (b-PS(+)) was correlated with poor outcomes (median PFS 1.5 months vs. 8 months in b-PS(-)) patients, p = 0.0007), while the presence of transversion mutations in KRAS and/or TP53 (b-KP-Tv(+)) predicted good outcomes (median PFS 11 months vs. 2 months in b-KP-Tv(-) patients, p = 0.0088). Combining these results, patients with a low “immune score” (driver and/or b-PS(+) and/or b-KP-Tv(-)) derived poor outcomes (PFS 2 months), compared with patients with a high immune score (no driver, b-PS(-) and b-KP-Tv(+), PFS 14 months, p = 0.0001, HR 2.96). Studying early changes in 65 specimens, molecular response was correlated with clinical outcomes (14 months PFS in patients with early ctDNA decrease compared to 2 months in patients with increase, p < 0.0001; HR 2.7). Using cut-off of 30% and 50% of plasma response increased the ability of ctDNA to predict response (HR 4 and 4.17, respectively). Conclusions: Targeted sequencing of plasma ctDNA and its early variations can predict response to anti-PD-1. Clinical trial information: NCT02827344