Inflammatory-nutritional biomarker scores and risk of malignant effusion in a pan-cancer cohort of 68,916 patients: Nonlinear associations and machine learning validation.

PURPOSE: Inflammatory-nutritional biomarker scores derived from routine blood tests have established prognostic value in cancer, yet their association with the occurrence of malignant effusions remains unexplored. This study systematically evaluated six composite scores: neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), lymphocyte-to-monocyte ratio (LMR), systemic immune-inflammation index (SII), prognostic nutritional index (PNI), and albumin-to-globulin ratio (AGR), for their nonlinear associations with cancer-related malignant effusion, and constructed an interpretable machine learning (ML) model to independently validate the role of the nutritional-immune axis. METHODS: This cross-sectional study included 68,916 cancer inpatients (2333 effusion events; 3.39%). Track 1 employed DAG-guided multivariable logistic regression and restricted cubic splines for nonlinear modeling; Track 2 built an XGBoost identification model via nested cross-validation with SHAP interpretability analysis. RESULTS: All six scores were independently associated with malignant effusion (all P < 0.0001). PNI exhibited the strongest protective association (adjusted OR 0.492 per SD increase) and the highest discrimination (AUC 0.697). Restricted cubic splines revealed significant nonlinear relationships with clinically actionable inflection points. The XGBoost model achieved AUROC 0.867; SHAP identified albumin and lymphocyte percentage, the core components of PNI and LMR, respectively, as top modifiable features, converging with Track 1 findings. CONCLUSION: Six blood-derived inflammatory-nutritional scores show independent, nonlinear associations with malignant effusion risk. Convergent statistical and ML evidence supports integrating these accessible biomarkers into early risk stratification, warranting prospective validation.