AIMS: Immune checkpoint inhibitors (ICIs) induce durable responses, but their clinical benefits apply to only a subset of patients. Therefore, precisely predicting a patient's response before ICI treatment is crucial. METHODS: A total of 248 patients with anti-Programmed cell death protein 1/Programmed death-ligand 1 (PD1/PD-L1)-treated advanced non-small cell lung cancer were enrolled, and clinical outcomes were collected with a minimum 6-month follow-up period. Tumour tissues were used for PD-L1 staining, targeted sequencing of 380 cancer-related genes and whole-exome sequencing (WES). RESULTS: The tumour mutation burden (TMB) obtained from targeted sequencing was higher among patients with a partial response (PR) than those with progressive disease (PD)/stable disease (SD) (P = 0.01) and in those with durable clinical benefit (DCB) than nondurable benefit (NDB) (P = 0.05). The somatic copy number alteration (SCNA) was lower in patients with a PR than those with PD/SD (P = 0.02) and in those with DCB than NDB (P = 0.02). The accuracy of the TMB and SCNA results from the targeted sequencing was confirmed by testing the correlation of the TMB and SCNA results from the targeted sequencing against those results from WES (r = 0.87, r = 0.62, respectively). To improve prediction score, TMB, SCNA and PD-L1 were integrated. New prediction scores reached Area under the ROC Curve (AUC) = 0.71 from TMB (AUC = 0.63), SCNA (AUC = 0.52) or PD-L1 (AUC = 0.57) with our cohort, and validation set from other cohorts also showed improved prediction scores with our new model. CONCLUSION: We report TMB, SCNA and PD-L1 as ICI biomarkers. Combining all these factors improved the prediction accuracy of ICI response compared with using individual factors. Tumour molecular features, TMB and SCNA, were efficiently obtained by targeted sequencing.