Development and validation of a mutation-based model to predict immunotherapeutic efficacy in NSCLC.

Background: Immunotherapy has become increasingly important in the perioperative period of non-small-cell lung cancer (NSCLC). In this study, we intended to develop a mutation-based model to predict the therapeutic effificacy of immune checkpoint inhibitors (ICIs) in patients with NSCLC. Methods: Random Forest (RF) classifiers were generated to identify tumor gene mutated features associated with immunotherapy outcomes. Then the best classifier with the highest accuracy served for the development of the predictive model. The correlations of some reported biomarkers with the model were analyzed, such as TMB, PD-(L)1, KEAP1-driven co-mutations, and immune subtypes. The training cohort and validation cohorts performed survival analyses to estimate the predictive efficiency independently. Results: An 18-gene set was selected using random forest (RF) classififiers. A predictive model was developed based on the number of mutant genes among the candidate genes, and patients were divided into the MT group (mutant gene ≥ 2) and WT group (mutant gene < 2). The MT group (N = 54) had better overall survival (OS) compared to the WT group (N = 290); the median OS was not reached vs. nine months (P < 0.0001, AUC = 0.73). The robust predictive performance was confifirmed in three validation cohorts, with an AUC of 0.70, 0.57, and 0.64 (P < 0.05). The MT group was characterized by high tumor neoantigen burden (TNB), increased immune infifiltration cells such as CD8 T and macrophage cells, and upregulated immune checkpoint molecules, suggesting potential biological advantages in ICIs therapy. Conclusions: The predictive model could precisely predict the immunotherapeutic efficacy in NSCLC based on the mutant genes within the model. Furthermore, some immune-related features and cell expression could support robust efficiency.