480P Genomic and immunohistochemical landscape of lung adenocarcinoma with concomitant tuberculosis: A two-stage whole-exome sequencing and validation study

Methods: Two independent cohorts of advanced NSCLC patients treated with IO or IO + chemotherapy at INT of Milan were analysed: APOLLO11 (n=91, training) and APOLLO (n=67, validation set).Peripheral blood was collected at baseline (T0) and first radiological evaluation (T1).Six FACS-derived features were considered: myeloid cells (CD11b+), neutrophils (CD66b+CD15+), mature/immature neutrophil ratio (CD10+/CD10-), and CD16+/CD16-ratios for mature, immature, and intermediate subsets.Two L2 Logistic Regression models were compared in predicting response (CR+PR vs. SD+PD): a baseline model (only T0 features), and a dynamic model (both T0 values and on-treatment changes as delta T1-T0).AUC was compared using the DeLong test, and SHAP analysis was used for explainability. Results:The baseline model achieved an external validation AUC of 0.526, while adding delta features significantly improved predictive performance (AUC=0.630,p=0.042).SHAP analysis identified baseline abundance of immature CD10-CD16neutrophils compared to CD10-CD16+ as poor predictors of response to IO.Conversely, a treatment-induced shift toward intermediate CD10intCD16+ and mature CD10+CD16+ phenotypes, compared to CD16-counterparts, predicted therapeutic benefit. Conclusions:Longitudinal neutrophil monitoring significantly improves baseline assessment in predicting IO response.Explainable ML identifies neutrophil maturation phenotypes as a biomarker of therapeutic benefit.These findings support the use of low-cost (300 ), minimally invasive FACS-based immune monitoring as an early decision-support tool to escalate treatments for a better IO response.