To investigate whether preoperative serum soluble PD-L1 (sPD-L1), chemokine ligand 9 (CXCL9), and spectral computed tomography (CT)-derived parameters could predict postoperative recurrence/metastasis in non–small cell lung cancer (NSCLC). A retrospective cohort of 200 NSCLC patients who underwent radical surgery between January 2019 and December 2021 was analyzed, with a follow-up period of 3 years. Patients were categorized into poor prognosis (recurrence/metastasis, n = 52) and good prognosis (n = 148) groups. A ridge-penalized logistic regression model was used, incorporating tumor size, tumor node metastasis (TNM) stage, lymph node metastasis, sPD-L1, CXCL9, effective atomic number (Eff-Z), normalized iodine concentration in the venous phase (NICVP), and normalized iodine concentration in the arterial phase (NICAP). Bootstrap optimism correction (1,000 resamples) and 10 × 10-fold cross-validation were applied for model validation, with performance evaluated via AUC, Brier score, and decision curve analysis (DCA). An independent external validation cohort was included to evaluate the model’s generalizability. The combined five-marker combined model yielded AUC of 0.904 (95% CI 0.860–0.949), significantly outperforming individual markers. Optimism-corrected AUC was 0.894, and calibration slope was 1.018. Brier scores were 0.107 (apparent) and 0.090 (corrected), indicating acceptable prediction error. DCA showed higher net benefit than “treat-all/treat-none” across 0.01–0.78 threshold probabilities. External validation showed that five-marker combined model maintained good discrimination and calibration, with an AUC of 0.824 (95%CI: 0.738–0.899), suggesting promising clinical applicability. Integrating circulating immune biomarkers with spectral CT iodine–based metrics provides promising performance for preoperative risk stratification of NSCLC recurrence/metastasis. Further temporal and multicenter validation is needed before clinical implementation.
Zhou et al. (Thu,) studied this question.