Cancer risk and mortality assessment in the overweight population: evaluation of single and integrated models based...

Chronic inflammation is a significant driver of carcinogenesis, particularly in the context of obesity; yet its role and the predictive value of common inflammatory biomarkers in overweight people (BMI 25-29.9) remain unclear and controversial. This study evaluated six hematologic inflammatory indices (BLR, NLR, MLR, SII, SIRI, PLR) for predicting cancer incidence and mortality among overweight adults.

Cross-sectional observational data from NHANES involving 11,348 overweight subjects and 9,181 normal-weight controls were analyzed; 823 overweight cancer patients were included for mortality assessment. Cancer status was self-reported, and cancer-specific death was defined by ICD-10 C00-C97 codes. Six inflammatory markers were derived from routine blood tests. Covariates were selected using LASSO and VIF, while nonlinearity was assessed through RCS. Predictive performance was evaluated using logistic regression, Cox proportional hazards models, and XGBoost with five-fold cross-validation.

All six markers presented U-shaped correlations with cancer risk and mortality. Fully adjusted models showed BLR independently predicted cancer risk (OR = 1.343, 95% CI:1.063-1.689, P = 0.012). NLR (HR = 1.150) and MLR (HR = 3.388) predicted mortality only in partially adjusted models, with significance lost after full covariate correction. Machine learning showed PLR_SII achieved the highest accuracy (0.7339) for cancer prediction, while PLR_SIRI yielded the maximum AUC (0.8124) among dual-marker panels, and MLR_SII_SIRI yielded optimal mortality forecasting (AUC = 0.7697, accuracy = 0.7273), with merely mild gains over single biomarkers. Subgroup disparities existed for MLR and SIRI across age, race and socioeconomic groups; elevated MLR/SIRI corresponded to worse survival per Kaplan-Meier curves.

Inflammatory markers have complex U-shaped links to cancer outcomes in overweight individuals. Combined multi-biomarker models deliver limited predictive improvement yet can aid population risk stratification. Given the observational design without external validation, only correlational conclusions can be drawn. Longitudinal prospective studies are required in the future to validate these indicators for clinical early cancer screening and intervention.