Integration of a severe immune-related adverse events predictive model with efficacy biomarkers enables precise stratification in NSCLC immunotherapy: insights from the phase 3 ORIENT-11 study

Abstract Background Severe immune-related adverse events (irAEs) compromise the outcomes of immune checkpoint inhibitors (ICIs). However, in routine practice, clinicians still lack biomarkers that can prospectively disentangle antitumor efficacy from the risk of severe irAEs, resulting in largely empirical, one-size-fits-all use of ICIs. Methods In this post hoc analysis of the randomized phase III ORIENT-11 trial, we analyzed RNA-sequencing data from baseline tumor samples of patients with advanced NSCLC receiving Sintilimab (an anti-PD-1 antibody) plus chemotherapy. GSEA and pathway enrichment analyses were performed to elucidate mechanisms of severe irAEs. A predictive model was constructed via 50-time LASSO regressions based on the enriched pathway and was subsequently combined with established efficacy biomarkers (PD-L1 expression and tumor-infiltrating lymphocyte [TIL]) to create a four-quadrant risk–benefit stratification system. Results Among 266 patients from the ORIENT-11 trial receiving combination therapy, severe irAEs occurred in 19 (7.14%). The Th17 cell differentiation pathway was significantly enriched in patients with severe irAEs. A predictive model derived from this pathway demonstrated AUC of 0.904 and 0.769 in training and validation cohort, respectively. Due to its independence of efficacy, integration of this toxicity model with PD-L1/TIL-defined efficacy biomarkers stratified patients into four groups, which exhibited differences in ORR (92.9% to 51.1%), severe irAE incidence (0% to 46.7%) and PFS. Conclusions The Th17 differentiation pathway is associated with severe irAEs for NSCLC patients. By integrating a Th17-based severe irAE prediction model with efficacy biomarkers, we propose a four-quadrant risk-benefit stratification framework that supports individualized immunotherapy decisions and moves beyond a one-size-fits-all approach.