Organ-specific progression and AI-based prognostic modeling in metastatic NSCLC treated with chemo-immunotherapy.

Abstract

e20585 Background: Organ-specific progression patterns and survival outcomes in metastatic non–small cell lung cancer (NSCLC) treated with chemo-immunotherapy are not well defined, and prognostic tools capable of predicting disease trajectories remain limited. This study evaluated organ-specific metastatic behavior, associated clinical outcomes, and machine-learning–based prognostic models in patients with advanced NSCLC. Methods: We retrospectively analyzed stage IV NSCLC patients without targetable mutations who received systemic chemo-immunotherapy at MD Anderson Cancer Center (2009–2025). Clinical and demographic variables, baseline metastatic sites, and first sites of progression were collected. Progression-free survival (PFS) and overall survival (OS) were assessed using Kaplan–Meier analyses. LASSO-Cox regression identified independent prognostic factors. Twelve machine-learning models were trained to predict disease progression and mortality, and model performance was evaluated using ROC curves, calibration, and SHAP-based feature interpretation. Results: Among 7,049 patients screened, 4,332 eligible stage IV patients were included. Progression most commonly occurred at the primary tumor (60.8%), lung (25.4%), bone (29.9%), brain (16.1%), liver (14.5%), and adrenal glands (12.7%). Liver progression was associated with the poorest outcomes (median PFS 5.07 months; median OS 11.80 months; P < 0.001). Patients with liver or bone progression experienced significantly shorter survival than those without involvement. Independent predictors of shorter PFS included ≥2 prior therapy lines, non-targetable EGFR mutations, adenocarcinoma, and baseline pleural metastasis. Poor OS was associated with baseline bone or liver metastasis, M3 stage, adenocarcinoma, and STK11 mutations. Of the twelve models evaluated, the ExtraTrees algorithm demonstrated the highest predictive accuracy for progression (AUC 0.935) and mortality (AUC 0.995). Conclusions: Organ-specific progression patterns strongly influence outcomes in metastatic NSCLC treated with chemo-immunotherapy, with liver progression conferring the worst prognosis. The high-performing, interpretable machine-learning models developed in this study may support personalized treatment planning for advanced NSCLC.